Deep Learning-Based Opportunistic Screening of Coronary Artery Disease on Non-Contrast Chest CT: A Multicenter Study

Coronary artery disease (CAD) is one of the leading causes of death worldwide. Many people have early atherosclerosis without symptoms, and some may develop significant coronary stenosis before any warning signs appear. Identifying high-risk individuals at an early stage is important to prevent heart attacks and other cardiovascular events.

Coronary CT angiography (CCTA) can directly evaluate plaque type and the degree of narrowing in the coronary arteries, but it is expensive, requires contrast injection, and involves higher radiation, making it unsuitable for large-scale screening. In contrast, non-contrast chest CT is widely used for health check-ups and lung disease follow-up. Such scans often provide clear views of certain coronary segments, which creates an opportunity to screen for CAD without additional cost or risk.

This multicenter study aims to develop and validate deep learning models to analyze coronary calcified segments that are visible on non-contrast chest CT. Two main objectives are: (1) to predict whether calcified segments contain mixed plaque components (both calcified and non-calcified); and (2) to predict whether these segments have significant narrowing (≥50% stenosis) as determined by CCTA. The study will also describe how often ≥50% stenosis is found in non-calcified segments, in order to demonstrate their low-risk nature.

The study includes retrospective data collected between 2015 and 2024, and a prospective external validation cohort starting in 2025. Approximately 1,417 patients with paired chest CT and CCTA have already been included for model development and testing. An additional 200 or more patients will be prospectively recruited for external validation.

This research may provide evidence that deep learning applied to routine non-contrast chest CT can serve as an opportunistic tool for early CAD risk screening in the general population.